OpenAI releases GPT-6.2 Astra and Sol; Astra's API price drops by 20 percent
The Decoder reports that OpenAI has made two new models available in ChatGPT and the API; independent measurements found no progress in coding.
According to a report by The Decoder dated 17 November 2026, OpenAI has released two new models, GPT-6.2 Astra and GPT-6.2 Sol. They are available on the Plus, Pro, Enterprise and Edu plans of ChatGPT and in the API. OpenAI says Astra's API price is 20 percent lower than its predecessor's, while Sol's price is unchanged.
According to a report by The Decoder dated 17 November 2026, OpenAI has released two new language models, GPT-6.2 Astra and GPT-6.2 Sol. The report says Astra is designed for specialist work in fields such as science, law and finance, while Sol is meant for quick, everyday tasks. The models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and developers can reach both through the OpenAI API.
- Astra pricing: 20 percent lower than its predecessor, according to OpenAI; 8 dollars per million input tokens and 40 dollars per million output tokens.
- Sol pricing: the same as GPT-6.1 Sol; 2 dollars per million input tokens and 10 dollars per million output tokens.
- Independent measurement: in the overall index of Artificial Analysis, Astra ranks second, behind Anthropic's Claude Fable 5.2.
- Caveats: according to the same organisation, Astra stayed level with its predecessor in the coding index and Sol dropped three points; the rate of guessing on hard factual questions was measured at 59 percent for Astra and 55 percent for Sol.
Why it matters
The report says both models reach the "Critical" threshold for cybersecurity in OpenAI's own risk framework, and the company is adding stricter safeguards for risky cybersecurity requests; OpenAI acknowledges these will also limit legitimate vulnerability research. Research organisation Epoch AI found Astra tied with Fable 5.2 on its own capabilities index but behind it on real research tasks.
For companies in Türkiye that use models through the API, the price change is a direct cost item: the per-token charge falls for applications running on Astra and stays the same for Sol. When choosing a model it makes sense to look beyond the results a vendor publishes, to independent measurements and trials on your own work samples; the caveats on coding and accuracy point the same way.
Sources
More articles
59 European tech founders and scientists ask the EU for a dedicated AI line in its 2028-2034 budget
The letter, published on 18 November 2026, warns that anything not secured in the EU's 2028-2034 budget talks will not exist before 2035.
November 18, 2026 · Source: EuronewsUS makes remote access to advanced AI chips through overseas data centers subject to export licences
As Nixon Peabody reports, the interim final rule BIS released on 16 November 2026 targets companies based in China renting computing power from data centers in third countries.
November 17, 2026 · Source: Nixon PeabodyNvidia reports 115.8 billion dollars in third-quarter revenue as Vera Rubin systems begin shipping
According to the company's release of 17 November 2026, most of the revenue came from data center products; it expects 130 billion dollars in the fourth quarter.
November 17, 2026 · Source: NVIDIA